private ip addresses from ISP

Robert Bonomi bonomi at
Tue May 23 14:47:01 UTC 2006

> Date: Tue, 23 May 2006 09:36:30 -0400
> To: nanog at
> From: Daniel Senie <dts at>
> Subject: Re: private ip addresses from ISP
> At 09:22 AM 5/23/2006, Robert Bonomi wrote:
> > > Date: Tue, 23 May 2006 03:33:34 -0400
> > > From: Richard A Steenbergen <ras at>
> > > To: nanog at
> > > Subject: Re: private ip addresses from ISP
> > >
> > >
> > > On Mon, May 22, 2006 at 04:30:37PM -0400, Andrew Kirch wrote:
> > > >
> > > > >   3) You are seeing packets with source IPs inside private space
> > > > > arriving at
> > > > > your interface from your ISP?
> > > ...
> > > > Sorry to dig this up from last week but I have to strongly disagree with
> > > > point #3.
> > > > >From RFC 1918
> > > >    Because private addresses have no global meaning, routing information
> > > >    about private networks shall not be propagated on inter-enterprise
> > > >    links, and packets with private source or destination addresses
> > > >    should not be forwarded across such links. Routers in networks not
> > > >    using private address space, especially those of Internet service
> > > >    providers, are expected to be configured to reject (filter out)
> > > >    routing information about private networks.
> > > >
> > > > The ISP shouldn't be "leaving" anything to the end-user, these packets
> > > > should be dropped as a matter of course, along with any routing
> > > > advertisements for RFC 1918 space(From #1). ISP's who leak 1918 space
> > > > into my network piss me off, and get irate phone calls for their
> > > > trouble.
> > >
> > > The section you quoted from RFC1918 specifically addresses routes, not
> > > packets.
> >
> >I quote, from the material cited above:
> >       "  ..., and packets with private source or destination addresses
> >        should not be forwarded across such links.  ...  "
> >
> >There are some  types of packets that can legitimately have RFC1918 source
> >addresses --  'TTL exceeded' for example -- that one should legitimately
> >allow across network boundaries.
> Really? You really want TTL-E messages with RFC1918 source addr? Even 
> if they're used as part of a denial of service attack? Even though 
> you can't tell where they actually came from?

"Can be" is not sufficient (in and of itself, that is) reason to block. 
_Anything_ "can be" used as part of a DOS attack.

TTL-E messages _do_ have legitimate function in network management.
TTL-E messages _can_ originate from RFC1918 space,  addressed to 'public
internet' addresses.  Usefully, and meaningfully.  Ever hear of 'traceroute'?
Ever use it where packets went across a network using RFC1918 internally?
Ever had a route die _between_ two RFC1918 addressed nodes on somebody elses

If you don't like that example, substitute "host/network unreachable", or
'ICMP redirect'. Or packet-size limit exceeded for a 'DNF' packet.  If you
don't get those messages back, you can't communicate.

> > >          If you're receiving RFC1918 *routes* from anyone, you need to
> > > thwack them over the head with a cluebat a couple of times until the cluey
> > > filling oozes out. If you're receiving RFC1918 sourced packets, for the
> > > most part you really shouldn't care.
> >
> >*I* care.
> >
> >When those packets contain 'malicious' content, for example.
> >
> >When the provider =cannot= tell me which of _their_own_customers_ originated
> >that attack, for example.  (This provider has inbound source-filtering on
> >their Internet 'gateway' routers, but *not* on their customer-facing 
> >equipment
> >(either inbound or outbound.)
> So you really don't want ANY packets with RFC 1918 source addresses 
> then, not even ICMP TTL-E messages, since they could be used in a 
> malicious fashion, and you would not be able to determine the true origin.

You need to learn to read.  I said "malicious content", not 'used in a
malicious fashion'.

Packets that could have legitimate meaning should be passed on.

Packets that _cannot_ have legitimate meaning should not be.

> >It's even more comical when the NSP uses RFC1918 space internally, and does
> >*not* filter those source addresses from their customers.
> You mean like Comcast using Cisco routers in their head-ends and 
> having the 10/8 address show up in traceroutes and so forth? 

not at all.  You're either ignorant of network architecture, or trying
to pick fights.

I was talking about a situation where a customer machine of a network
that uses RFC1918 addresses internally starts sending malicious packets
with a RFC1918 source address that _matches_ that of one of the *in*use*
addresses on the service-provider network.  AND the service-provider does
not do ingress (from the customer) or egress (to the customer) filtering
of RFC1918 address-space.  Customer A's machine starts attacking Customer 
B, C, D, E, F ...;  those ill-informed customers don't null-route outbound 
RFC1918 addresses; you get an 'inadvertent' smurf attack on the NSP resource.
It's comical, because the ISP's 'bad practice' facilitated the attack on
the ISP.

"Traceroute", by the way, *is* one of those 'legitimate' cases for which 
RFC1918 source-address packets should be allowed across network boundaries.

Proper "good net neighbor" egress filtering of RFC1918 source addresses 
takes a number of separate rules.  Several 'allows', followed by a default

>                                                              Not sure 
> to what degree it's the NSP's fault vs. the router vendors', but yes.

Can't imagine why you think it might be the fault of the router vendor.

Can't imagine  why think it is somebody's "fault".

> > >                                      There are semi-legitimate reasons for
> > > packets with those sources addresses to float around the Internet, and
> > > they don't hurt anything.
> >
> >I guess you don't mind paying for transit of packets that _cannot_possibly_
> >have any legitimate purpose on your network.
> Along with this goes the usual flamewar over RFC 2827, ingress 
> filtering (of which URPF is a subset implementation).

If everybody did _egress_ filtering of 'cannot possibly be legitimate' traffic,
ingress filtering of that traffic would not be necessary.  Unfortunately, not
everybody does, so ingress filtering _is_ necessary.

'ingress' filtering is self-defense.
'egress' filtering is about 'being a good net neighbor'.

> >Some of us, on the other hand, _do_ object.
> And some of us pay for bandwidth, care about getting congestion 
> problems from useless traffic, etc. Perhaps it makes the case a lot 
> clearer for selling "better than equal" service to the highest bidder 
> if your network is overrun with undesired traffic.

More information about the NANOG mailing list